How the text comparison works
The tool reads the text of every page of both files with pdf.js, in reading order, and compares the two documents word by word with the Myers difference algorithm (E. W. Myers, 1986), which finds the smallest set of words to remove and add to turn one text into the other. It compares the whole document at once rather than page against page, so when an inserted paragraph pushes the rest of the text down onto the next page, only the new paragraph is reported. Spacing and line breaks are ignored, because the same sentence can wrap differently in two versions of a PDF. Each removed word is shown where it used to be, between the words around it, so a changed amount reads as the old figure struck through, followed by the new one.
Visual comparison for scans, stamps and layout
Some changes are not in the text at all: a new signature, a stamp, a swapped logo, a moved table or a different scanned page. With the visual check on, each page of the changed file is drawn at the same width as the matching page of the original, and every pixel that changed is counted. Choose a page to see the overlay: the changed page in light grey, ink that is only in the original in red, ink that is only in the changed file in green, and ink that changed colour in amber. Pages are matched by their text, so an inserted page does not throw the following pages out of step; pages with no text at all, such as scans, are matched by their position. Two PDFs made by different programs can differ in pixels even when the words are identical, because the fonts are drawn differently; in that case, trust the text result first.
Arabic and right-to-left documents
Arabic lines are read right to left, with English names, IBANs and numbers inside them kept in their own order, and joined letter shapes are turned back into ordinary letters before comparing, so two copies of the same Arabic contract match even when they were made by different programs. Arabic words are compared as whole words with their marks; tick "Ignore Arabic marks" when one version has tashkeel and the other does not. In the results and in the report each line takes its own direction, so Arabic lines read from the right and English lines from the left on the same page.
Different page counts, scanned pages and protected files
The two files do not need the same number of pages. Pages that only the changed file has are listed as new pages, and pages of the original whose text is completely gone are listed as removed pages, in their place in the document. A scanned page has no text, so it can only be compared visually, and the tool lists such pages for each file. To compare the words of scans, read them first with Image to Text (OCR) and compare the two texts with Text compare. A password-protected PDF must be unlocked first with Unlock PDF, and damaged or empty files get a clear message.
The report and your privacy
The report is a single HTML file that opens in any browser, on a phone too, and can be printed or saved as a PDF from there. It holds the summary, every changed page with the added and removed words marked, and the visual overlays of the pages that look different. Both PDFs are read by this browser tab and never uploaded, which matters for contracts, tenders and HR letters. There is no sign-up and no limit on how many files you compare.